| | |
| | | layer->bias_updates = calloc(n, sizeof(float)); |
| | | layer->bias_momentum = calloc(n, sizeof(float)); |
| | | float scale = 1./(size*size*c); |
| | | scale = .01; |
| | | scale = .05; |
| | | for(i = 0; i < c*n*size*size; ++i) layer->filters[i] = scale*2*(rand_uniform()-.5); |
| | | for(i = 0; i < n; ++i){ |
| | | //layer->biases[i] = rand_normal()*scale + scale; |
| | |
| | | |
| | | const size_t global_size[] = {layer.n}; |
| | | |
| | | clEnqueueNDRangeKernel(queue, kernel, 1, 0, global_size, 0, 0, 0, 0); |
| | | cl.error = clEnqueueNDRangeKernel(queue, kernel, 1, 0, global_size, 0, 0, 0, 0); |
| | | check_error(cl); |
| | | } |
| | | |
| | |
| | | |
| | | const size_t global_size[] = {layer.n*size, layer.batch}; |
| | | |
| | | clEnqueueNDRangeKernel(queue, kernel, 2, 0, global_size, 0, 0, 0, 0); |
| | | cl.error = clEnqueueNDRangeKernel(queue, kernel, 2, 0, global_size, 0, 0, 0, 0); |
| | | check_error(cl); |
| | | } |
| | | |
| | |
| | | |
| | | gemm_ongpu_offset(0,1,m,n,k,1,a,i*m*k,k,b,i*k*n,k,1,c,0,n); |
| | | } |
| | | //cl_read_array(layer.delta_cl, layer.delta, m*k*layer.batch); |
| | | |
| | | if(delta_cl){ |
| | | m = layer.size*layer.size*layer.c; |